I just came across this 12 minute talk about how to evaluate voting methods. I pretty much agree with it, but what do you think?
-
Features
Evaluating voting methods
- Thread starter Blahface
- Start date
fast
Contributor
Interesting.
It basically says that you can arbitrarily decide you like a voting system because it passes a certain criteria such as “clone independence”, “later no harm,” but the best way to evaluate a voting system is to test it out through many different simulations and calculate the Bayesian regret. When you do this, range voting and approval voting come out on top.
Duke Leto
Deleted
It looks like the "Bayesian Regret" is actually "the difference between a person's range voting optimal candidate and the range voting winner", when subjected to a bunch of test elections with made up but "realistic" ballots.
So it's not so much that the voting methods are being evaluated independently and range and approval are coming out on top, it's that you're making the assumption that a range ballot most accurately captures voter intent, which is probably not a good assumption. (It only works if the voter accurately rates the degree of unacceptability or acceptability between sets of candidates. What would be my relative rankings on the mostly identical Green Presidential Candidate Jill Stein and Justice Party's Rocky Anderson?) Circular reasoning, basically.
So it's not so much that the voting methods are being evaluated independently and range and approval are coming out on top, it's that you're making the assumption that a range ballot most accurately captures voter intent, which is probably not a good assumption. (It only works if the voter accurately rates the degree of unacceptability or acceptability between sets of candidates. What would be my relative rankings on the mostly identical Green Presidential Candidate Jill Stein and Justice Party's Rocky Anderson?) Circular reasoning, basically.
lpetrich
Contributor
Voting And Election Reform has Simulation Of Various Voting Models for Close Elections
The voting algorithms:
Changing the number of candidates in the absence of noise, FPTP quickly gets worse than the others, being about 1/6 as good as the others for 40 candidates. IRV is also worse than the others, but not by much.
Rated-vote systems were the best in both cases, on average beating Condorcet, though not by much. Approval voting was the worst of such systems, doing a little worse than Condorcet.
So although IRV is better than FPTP, it is not as good as it could be.
The voters' preference values are generated randomly, using the C library function random(). In many of the runs, voter misinformation and confusion were simulated by adding random numbers to the preference values before calculating votes from them.
The voting algorithms:
- One Vote (Plurality Voting for First Past The Post) -- most preferred candidate gets 1, all others get 0
- Acceptance Voting (Approval Voting) -- candidates with positive preferences get 1, all others get 0
- Rated Vote -- use the voters' internal model -- equal to Maximum Happiness
- Rated Vote, equal sum -- adjust the preferences so that they add up to 1
- Rated Vote, maximized -- scale the preference so that the maximum by absolute value is +1 or -1
- Rated Vote, 1...N -- scale preferences into 1 to #, then quantize to integer values
- Rated Vote, 1...10 -- like above, but 1 to 10
- Ranked Voting (Borda count) -- make highest preference N, next highest N-1, down to 1
- Ranked Voting, no negative preferences
- Instant Runoff Voting (Single Transferable Vote)
- IRNR -- rather complicated
- Condorcet
- Random
Changing the number of candidates in the absence of noise, FPTP quickly gets worse than the others, being about 1/6 as good as the others for 40 candidates. IRV is also worse than the others, but not by much.
Rated-vote systems were the best in both cases, on average beating Condorcet, though not by much. Approval voting was the worst of such systems, doing a little worse than Condorcet.
So although IRV is better than FPTP, it is not as good as it could be.
lpetrich
Contributor
Finest governmental system? - YouTube Addresses the claim in the John Birch Society's documentary "Overview of America" that the US has the "finest governmental system ever devised by man". The author of that video then looked at how different nations fare in The Economist magazine's Democracy Index. That magazine's Intelligence Unit has published ratings 6 times since 2007, and the ratings cover:
I Electoral process and pluralism, II Functioning of government, III Political participation, IV Political culture, V Civil liberties
In that video, the US was the only Presidential Republic in the top 20. Switzerland was a Direct Democracy (???) and the rest were Parliamentary Democracies. Of the 10 least democratic countries, 5 were Presidential Republics. They have such company as Saudi Arabia and North Korea, the worst of them all.
The video then quoted a US politician stating "This is not a parliament." Followed by another politician stating "We are going to bring the President here and have a question period? Like the Prime Minister has in Great Britain?" That was likely meant as sarcasm, but that seems like a good idea.
That video was made in 2009, so it will be a bit out-of-date. But I've redone those comparisons, and I find that that video's claims are still mostly correct.
I found that parliamentary republics and quasi-republics were the top scorers. Quasi-republics? That's to include figurehead monarchies, because they act like republics most of the time. Presidential republics don't score as high. With the most recent numbers, I find that the US is ranked 22 with 8.11 out of 10, with Uruguay and South Korea a bit ahead, and Costa Rica, Portugal, and France a bit behind.
I also found that party-list proportional representation and similar systems were the favorite systems for electing legislatures in the top-scoring countries. Single-member-district first-past-the-post elections were used by only a few high scorers; the only ones that scored higher than the US were Canada and the UK.
So one might conclude from this that the best system of government that humanity has ever devised is the parliamentary republic with its legislature elected by proportional representation.
I Electoral process and pluralism, II Functioning of government, III Political participation, IV Political culture, V Civil liberties
In that video, the US was the only Presidential Republic in the top 20. Switzerland was a Direct Democracy (???) and the rest were Parliamentary Democracies. Of the 10 least democratic countries, 5 were Presidential Republics. They have such company as Saudi Arabia and North Korea, the worst of them all.
The video then quoted a US politician stating "This is not a parliament." Followed by another politician stating "We are going to bring the President here and have a question period? Like the Prime Minister has in Great Britain?" That was likely meant as sarcasm, but that seems like a good idea.
That video was made in 2009, so it will be a bit out-of-date. But I've redone those comparisons, and I find that that video's claims are still mostly correct.
I found that parliamentary republics and quasi-republics were the top scorers. Quasi-republics? That's to include figurehead monarchies, because they act like republics most of the time. Presidential republics don't score as high. With the most recent numbers, I find that the US is ranked 22 with 8.11 out of 10, with Uruguay and South Korea a bit ahead, and Costa Rica, Portugal, and France a bit behind.
I also found that party-list proportional representation and similar systems were the favorite systems for electing legislatures in the top-scoring countries. Single-member-district first-past-the-post elections were used by only a few high scorers; the only ones that scored higher than the US were Canada and the UK.
So one might conclude from this that the best system of government that humanity has ever devised is the parliamentary republic with its legislature elected by proportional representation.
I'm not sure how they calculate scores based on their utility values for each candidate, but it is not a direct correlation. According to the explanation on their site:
It is hard to determine. There are so many systems that could be tried, but are untested.
Duke Leto
Deleted
I'm inherently suspicious of a group explicitly advocating Range Voting finding and showcasing "statistical proof" of the superiority of Range Voting.
lpetrich
Contributor
Maybe, but one could confine one's attention to existing systems.
One need not depend on The Economist's rankings. One finds much the same results for The Fund for Peace's Fragile-State rankings, formerly the Failed-State rankings. Somalia, the longtime winner, is now in second place, displaced by South Sudan. The least fragile state in this assessment is Finland, followed by other northern-European and overseas-European nations. They are all parliamentary republics or quasi-republics. The first presidential republic at that end of the list is France, and it's often called a semi-presidential one (President + Prime Minister). The US is right behind it, at #20.
ClayShentrup
New member
This is a common misunderstanding. One obvious indication it's wrong is that Score Voting (aka Range Voting) didn't have a perfect Bayesian regret of zero in the graph.
Bayesian regret is measured from actual utilities, which are accurately known, because we make them up.
Scores, on the other hand, are a lossy transform (a "blurring") of actual utilities. The loss comes from:
- Normalization
- Ignorance (e.g. I think President Smith will be great, but he ends up wiping out my ethnic group)
- Tactical behavior (I discussed this in the video)
I do wish people would take the time to read about a complex subject like this before making such arguments. I've had to explain this numerous times over the years.
Clay Shentrup
ClayShentrup
New member
A parliamentary legislature is a completely separate topic from proportional representation. A parliamentary system basically just means that the legislature elects the executive.
https://en.wikipedia.org/wiki/Parliamentary_system
As for the quality of proportional representation, this is a very complex subject, and much intuitive thinking is wrong.
http://scorevoting.net/PropRep.html
One major point there is that PR is irrelevant to single-winner offices like mayor/governor/senator/president/etc.
Another is that if you use proportional representation, there are better PR systems, such as Proportional Score Voting or Asset Voting. The world currently uses antiquated schemes like STV and MMP, which were not designed by experts. (Well, maybe they were experts in their day, but social choice theory has hugely advanced in the past few decades.)
Lastly, we have also never seen any country using a GOOD single-winner system, like score voting or approval voting. Most use plurality and some use IRV, both of which are poor. So it's not clear that PR really is the best approach.
Duke Leto
Deleted
I'm also inherently skeptical of people with the free time and enthusiasm to come onto TalkFreethought to defend their positions when not actually summoned by anyone and who appear therefore to be using some sort of Google Alert to pop up anywhere on the web their work is questioned. Apologies if one of the posters emailed you asking you to weigh in, but people watching the web for any contradiction of their position anywhere are generally ones to run away from very fast.
I apologize for the previous misunderstanding on the mathematical modelling of regret, but it still looks like there's plenty of room to haggle "shampoo versus conditioner", and it's not as big a slam dunk as you are representing it. Here are the issues I see with the study (or studies if this based on more than just the Warren Smith paper).
First, the phrase "because we made them up" does not inspire confidence, because it means the test is only potentially as accurate as the model of an electorate you make before the blurring. Real aggregates of people might have patterns of preferences different from the expectations of pure theory. If you are not familiar with Adrian Thompson's work in evolving circuitry designs using physical re-configurable circuit boards, the moral is that he ended up with a circuit in which one of the vital components is not connected to main circuit system, and interacts with the main system to produce the desired behavior in apparent contravention of the most basic law of circuit design, and which would not have been discovered had the circuits just been modeled by software using the standard rules of circuit design. Reality has an irritating habit of throwing things we don't theoretically expect at us.
Second, the video indicates that the test was done on multiple iterations of five candidate sets, and I would like to see if the result continues to appear as definite as the number and ideological diversity of the candidates increases. One thing that has always struck me about the classic "Tennessee Capital" thought experiment is that the number of options (generally four) may be as much a problem as the winner determination method itself. Two candidates are clumped on the "far right" with one in the "center right" and one on the "far left". What happens as more minor population centers get pulled out and added to the election as candidates? Does the disconnect between the Condorcet (and Range) winner and the IRV winner persist as Clarksville and Murfreesburo enter the race? Does Range's superiority persist with 7 or 10 candidates? And so on. One of the presumed shared objectives of everyone who dislikes FPTP is desire for greater candidate diversity, so are we sure that the possible replacements will be ideal at any amount of candidate diversity, not just the amount we think is likely?
Third, and this is intuitively the strongest theoretical objection, I think, did the test accurately blur in response to differential voter exhaustion? Absolute ranking on a scale takes slightly more mental effort than preference ranking, and moreso as the number of candidates goes up. Asking voters to do more work in this way may actually be source of diminishing returns.
The other objections I have are practical rather than theoretical. I'm still in favor of IRV over Condorcet and Range, although this discussion has moved Range ahead of Condorcet for me. Most voters are mathematically illiterate, and the one metric on which FPTP beats all other contenders is its comprehensibility to laymen. It may be that it is mathematically demonstrable that the more complex winner picking systems yield more desirable results, but how many voters will understand that when it's explained to them? Of those that do not understand, how many will believe the assurances of mathematicians that such and such a system is better? If most voters don't accept the system as legitimate, you're stuck in endless accusations of fraud and conspiracy, which is probably not preferable to the FPTP Duopoly, because Major Parties DO collapse completely from 3rd Party challenges from time to time. Range does beat Condorcet on the "can be explained to a moron" metric, but it doesn't beat out IRV because the algorithm is simpler.
More importantly, there's the problem of implementation, especially in the USA, because you need to get the approval of the same politicians whose re-election will be endangered by any change to the status quo. The comparative lack of fitness for IRV in eliminating the Duopoly may actually be a feature rather than a bug from that standpoint as far as the US is concerned.
Empirically, I'd be happiest with a nationwide exit poll that takes preferences from actual voters of candidates that are in the actual races at the same time as elections, and then tracks their satisfaction in follow up interviews on preferences after the "alternate winners" are calculated, and one that does this over a few cycles. The problem here is in getting an accurate sample. It'd also be horrifically expensive.
I hope this makes some of my misgivings clearer.
I agree that we need to reduce voter exhaustion. I think though that rating candidates is more intuitive than ranking them. I also think that at the moment, even rating is asking for too much from the voter and we shouldn't start with that as our first attempt at reform. Another fear I have with range voting is that different groups might have different methods of rating. Maybe some people would just decide to rate unknown candidates in the middle and some other people would automatically give unknown candidates a zero rating.
I think the best solution is to have a non-partisan primary with approval voting and allow the general election to be a fine tuning of the primary in which the top two would face off.
If you are worried about confusion among voters, IRV is one of the worst in that regard. I think it is simple to understand, but apparently I'm wrong. Even an election official in Oakland couldn't explain how it works.
What you describe about people not trusting a system that they don't understand – that actually happened in Burlington, Vermont when they implemented IRV. The Progressive Bob Kiss won the mayoral election even though he would have lost head to head with the Democrat. People thought they were being cheated and were pissed off so they voted to repeal IRV.
Another problem with IRV is that it is not additive. Ballots have to be recounted each round. You also can't restrict a recount to specific area that has a high spoiler rate for ballots like what Gore tried to do in 2000. A change in who gets eliminated first in the early rounds can necessitate a complete recount. Condorcet methods don't have this problem.
Edit:
I'm not sure why you think this is a bad thing for the original author to clarify his work on a thread that created specifically to discuss it.
Last edited:
lpetrich
Contributor
When has that happened?
There have been two major-party collapses in US history, and neither of them seem provoked by third parties, at least to me. The Federalists around 1820 and the Whigs around 1860. The Federalists were succeeded by the Whigs, and the Whigs by the Republicans.
lpetrich
Contributor
Like what?
Because it is not applicable to single-seat elections.
I'd like to see some explanations of those systems.
Several countries use a two-ballot top-two runoff system, like France. Is that just as bad?
Not being applicable to single-seat elections is not the same thing as being deficient for multiseat ones.
Duke Leto
Deleted
We're sufficiently minor that it implies either someone asked him to come by or that he's taking it way too seriously. I had a really bad experience with this happening on one of the previous iterations of FRDB, although to be fair this was about biblical commentary, which attracts the insane.
Duke Leto
Deleted
Getting a good uniform method of casting ballots that ensures people are given their voting rights and making the process auditable is almost as big a problem as getting the winner determination right. We should probably confine the thread to one or the other.
If this Bob Kiss is who I think it is, as in the VoteKiss party, then the correct description of him would not be "Progressive" but "Completely Insane".
Interesting example though.
Duke Leto
Deleted
Labour taking down the Liberals in the UK. I'm less familiar with the rest of the commonwealth countries, but I understood that this was the way the change went in many of them. In Australia, that the "Liberals" are on the right implies the ancestral Tories just curled up and died to make room for Labour. I think NDP has just pulled off one in Canada against the Liberals, but that may reflect a different election system.
"Once every two centuries per country" is technically more frequent than "Never".